The Agent Skills Directory

[DYNAMIC_CONTEXT_INJECTION]: The SKILL.md file contains dynamic context injection placeholders (! commands) that execute gcloud config get-value project and gcloud auth list at load time. These commands are used to automatically populate the agent's context with the current project ID and active account to facilitate project-specific workflows.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to analyze external data such as agent configurations and source code, which introduces a potential surface for indirect prompt injection.
Ingestion points: Deployment configurations retrieved from the Vertex AI API (reasoningEngines.get) and agent source code files accessed via Read, Glob, and Grep for linting and security scanning in SKILL.md and references/security-checklist.md.
Boundary markers: The instructions do not explicitly define delimiters or specific "ignore" instructions for the agent to distinguish between its own logic and potentially malicious instructions embedded in the data being audited.
Capability inventory: The skill has access to powerful tools including Bash for gcloud, python, and security scanners like bandit, which could be targeted by successful injection.
Sanitization: There is no evidence of specific input sanitization or filtering logic applied to the audited configurations before they are processed for scoring and remediation.

validator-expert